> why we need an integrator in DC servo instead of just an error correction amplifier?
As far as the "error correction" is concerned, our audio is an "error". It would try to reduce this error to zero. No audio output.
What we really want is: DC is an error, audio is not an error. The simplest way to define what is DC and what is audio is with a low-pass filter. And the most extreme low-pass is the integrator, because it tends to infinitely high gain (error reduction) at infinitely low frequency (DC).
> How to implement DC servo/offset compensation in a fully balance system?
You have errors in two terms, you need two servos. There may be some very clever way to share parts between two servos, but it still has at least two inputs, two outputs, and two gain functions. If you are not going into mass production, two or even three servos is the most direct and maybe-successful way to approach it.
Thanks for your reply. Do you have any examples which have offset cancellation in full balance amplifier? So far I have only seen single ended version.
My concern for offset cancellation in full balance design is to cancel any offset cause by device mismatch in differential amplifier (more severe if high gain), where the usual common-mode feedback (CMFB) is unable to correct. Hope you can understand my desciption.
current mirror loading assures Ic balance to within a couple of Ib, easily better than 1% Ic match, this should give sub mV input refered offset V - with well matched input Qs
In DIY you have the luxury of discounting labor in favor of craftmanship so matching input transistors should be expected
Manual trimming which is anathema to the production engineer is another DIY option; small emitter degeneration R can be trimmed for 0 offset V somewhat independant of Ic balance (which can/should be tirmmed depending on topology)